Research map / Large language models / Instruction and fine-tuning
TermGPT: Multi-Level Contrastive Fine-Tuning for Terminology Adaptation in Legal and Financial Domains
Yidan Sun, Mengying Zhu, Feiyue Chen, Yangyang Wu, Xiaolei Dan, Mengyuan Yang, Xiaolin Zheng, Shenglin Ben
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add SFT stage before contrastive learning (ablation) | TermGPT-Qwen3 · JecQA QCA | accuracy: 0.796 -> 0.858 | Improved |
| add SFT stage before contrastive learning (ablation) | TermGPT-Qwen3 · Financial Regulations QCA | accuracy: 0.818 -> 0.908 | Improved |
| add contrastive learning on top of SFT (ablation) | TermGPT-Qwen3 · JecQA QCA | accuracy: 0.834 -> 0.858 | Improved |
| add contrastive learning on top of SFT (ablation) | TermGPT-Qwen3 · Financial Regulations QCA | accuracy: 0.864 -> 0.908 | Improved |
| add token-level contrastive learning (ablation) | TermGPT-Qwen3 · JecQA QCA | accuracy: 0.844 -> 0.858 | Improved |
| add token-level contrastive learning (ablation) | TermGPT-Qwen3 · Financial Regulations QCA | accuracy: 0.899 -> 0.908 | Improved |
| add sentence-level contrastive learning (ablation) | TermGPT-Qwen3 · JecQA QCA | accuracy: 0.849 -> 0.858 | Improved |
| add sentence-level contrastive learning (ablation) | TermGPT-Qwen3 · Financial Regulations QCA | accuracy: 0.894 -> 0.908 | Improved |
| use Qwen3-8B instead of LLaMA-8B backbone (ablation) | TermGPT · JecQA QCA | accuracy: 0.681 -> 0.858 | Improved |
| use Qwen3-8B instead of LLaMA-8B backbone (ablation) | TermGPT · Financial Regulations QCA | accuracy: 0.856 -> 0.908 | Improved |
| TermGPT multi-level CL vs best 8B contrastive baseline | Qwen3-8B · JecQA QCA | accuracy: 0.834 -> 0.858 | Improved |
| TermGPT vs commercial LLMs | TermGPT-Qwen3 · JecQA QCA | accuracy: 0.894 -> 0.858 | Hurt |
Closest papers by meaning
- SaulLM-54B & SaulLM-141B: Scaling Up Domain Adaptation for the Legal Domain (NeurIPS 2024 · similarity 0.84)
- The SEC Filings Dataset: Reconstructing U.S. Corporate and Financial Disclosures into Layout-Faithful and Token-Efficient Pretraining Data (NeurIPS 2026 · similarity 0.83)
- DELTA: Pre-Train a Discriminative Encoder for Legal Case Retrieval via Structural Word Alignment (AAAI 2025 · similarity 0.83)
- Lawma: The Power of Specialization for Legal Annotation (ICLR 2025 · similarity 0.83)
- FinBen: A Holistic Financial Benchmark for Large Language Models (NeurIPS 2024 · similarity 0.83)
- Rethinking the Development of Large Language Models from the Causal Perspective: A Legal Text Prediction Case Study (AAAI 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
